Hardware implementation aspects of polar decoders and ultra high-speed LDPC decoders

The goal of channel coding is to detect and correct errors that appear during the transmission of information. In the past few decades, channel coding has become an integral part of most communications standards as it improves the energy-efficiency of transceivers manyfold while only requiring a modest investment in terms of the required digital signal processing capabilities. The most commonly used channel codes in modern standards are low-density parity-check (LDPC) codes and Turbo codes, which were the first two types of codes to approach the capacity of several channels while still being practically implementable in hardware. The decoding algorithms for LDPC codes, in particular, are highly parallelizable and suitable for high-throughput applications. A new class of channel codes, called polar codes, was introduced recently. Polar codes have an explicit construction and low-complexity encoding and successive cancellation (SC) decoding algorithms. Moreover, polar codes are provably capacity achieving over a wide range of channels, making them very attractive from a theoretical perspective. Unfortunately, polar codes under standard SC decoding cannot compete with the LDPC and Turbo codes that are used in current standards in terms of their error-correcting performance. For this reason, several improved SC-based decoding algorithms have been introduced. The most prominent SC-based decoding algorithm is the successive cancellation list (SCL) decoding algorithm, which is powerful enough to approach the error-correcting performance of LDPC codes. The original SCL decoding algorithm was described in an arithmetic domain that is not well-suited for hardware implementations and is not clear how an efficient SCL decoder architecture can be implemented. To this end, in this thesis, we re-formulate the SCL decoding algorithm in two distinct arithmetic domains, we describe efficient hardware architectures to implement the resulting SCL decoders, and we compare the decoders with existing LDPC and Turbo decoders in terms of their error-correcting performance and their implementation efficiency. Due to the ongoing technology scaling, the feature sizes of integrated circuits keep shrinking at a remarkable pace. As transistors and memory cells keep shrinking, it becomes increasingly difficult and costly (in terms of both area and power) to ensure that the implemented digital circuits always operate correctly. Thus, manufactured digital signal processing circuits, including channel decoder circuits, may not always operate correctly. Instead of discarding these faulty dies or using costly circuit-level fault mitigation mechanisms, an alternative approach is to try to live with certain malfunctions, provided that the algorithm implemented by the circuit is sufficiently fault-tolerant. In this spirit, in this thesis we examine decoding of polar codes and LDPC codes under the assumption that the memories that are used within the decoders are not fully reliable. We show that, in both cases, there is inherent fault-tolerance and we also propose some methods to reduce the effect of memory faults on the error-correcting performance of the considered decoders.


  • Thesis submitted - Forthcoming publication

Related material